Single-Ended vs Differential Signaling: Key Differences Explained

Single-ended signaling reads the voltage on one signal line relative to ground, which is simple and needs few wires. Differential signaling sends opposite signals on two lines and reads 0 and 1 from the voltage difference between them. Noise and ground potential differences that appear equally on both lines cancel when the difference is taken, and the electromagnetic fields from the opposite currents also cancel, so signals can be sent faster and farther with a small swing. The trade-off is that each signal needs two lines, and the design must match the length and spacing of the two lines.

AspectSingle-endedDifferential
Wires per signalOne (the return current shares ground)Two (a positive and negative pair)
Reference for readingVoltage relative to groundVoltage difference between the two lines
Common-mode noiseAdds directly to the signalCancels when the difference is taken (depending on the receiver's CMRR)
Ground potential differencesA ground offset between transmitter and receiver becomes error directlyLittle effect as long as it is within the receiver's allowed common-mode voltage range
Radiated noise (EMI)Current loops are large, so it radiates easilyFields from the opposite currents cancel, so it radiates little
Signal swingLarge. Close to the supply voltage (e.g. 0 to 3.3 V)Small. A few hundred mV (about 350 mV for LVDS)
Typical speedTens to hundreds of Mbit/s. Memory buses with careful termination reach several Gbit/s per lineHundreds of Mbit/s to tens of Gbit/s per pair
Routing designSimple. For fast signals, watch the characteristic impedance and terminationMatch the length and spacing of the two lines and maintain the differential impedance (often 85 to 100 Ω)
Cost and pin countLow cost, fewer pinsTwice the pins and wiring. Transmitter and receiver circuits are also more complex
Typical examplesGPIO, I2C, SPI, UART, DDR data (DQ)LVDS, USB, PCI Express, SATA, HDMI, Ethernet, CAN, RS-485

Figures are typical orders of magnitude and vary by process generation and product.

When to use which

Single-ended
When speed is not very high, distances are short, and you want to keep pin count and cost down. On-board GPIO, I2C, SPI, etc.
Differential
For high-speed signals in the Gbit/s class, connecting separate equipment by cable, noisy environments, or when radiated noise must be suppressed. PCIe, USB, Ethernet, automotive CAN, etc.
Example of using both
DDR memory uses single-ended signaling for the numerous data lines (DQ) and differential signaling for the clock and strobes (DQS) that serve as timing references, balancing pin count and reliability.

FAQ

Why is differential signaling robust against common-mode noise?

When two lines run close together, external noise couples onto both by nearly the same amount. The receiver reads only the voltage difference between them, so noise that appears equally on both (common-mode noise) is removed by subtraction. A shift in ground potential also appears equally on both lines, so it has little effect.

Differential signaling needs two lines, so why can it go faster?

Because it is robust against noise, the swing can be reduced to a few hundred mV, and the smaller voltage to charge and discharge allows faster switching. It also radiates little and is less sensitive to ground potential differences, so serial links can reach tens of Gbit/s per pair.

What should you watch for when routing a differential pair?

First, match the lengths of the two lines. If the lengths differ, the two signals arrive at different times, the difference waveform is distorted, part of it turns into common-mode noise, and radiation increases. Also keep the spacing constant to hold the differential impedance at the specified value (often 85 to 100 Ω), and terminate at the receiver.

Does differential signaling need no ground?

It is not that no ground is needed at all. The average of the two voltages (the common-mode voltage) must stay within the range the receiver allows. To ensure this, either share a common ground, or block DC with capacitors or transformers and re-establish the common-mode voltage at the receiver. Ethernet uses transformers to isolate the equipment from each other.

Check your understanding

Why do high-speed serial links use differential signaling, sending opposite signals on two wires?

  1. ATwo wires always double the amount of data sent
  2. BIf one wire breaks, the other alone carries the data at the same speed
  3. CNoise picked up equally on both wires cancels when the difference is taken, so small swings can be sent fast and reliably
Show answer and explanation

Answer: C. Noise picked up equally on both wires cancels when the difference is taken, so small swings can be sent fast and reliably

In differential signaling, 0 and 1 are represented by the voltage difference between two wires. External noise couples almost equally onto the two closely routed wires, so it cancels when the difference is taken. This allows swings of only a few hundred millivolts, reducing the time and power needed to change voltage and enabling high speed. Because the two currents flow in opposite directions, electromagnetic emissions are also small.

Why have high-speed connections between chips and boards moved from parallel transmission over many wires to serial transmission over a few fast wires?

  1. AIn parallel transmission, arrival-time differences between wires (skew) make it hard to raise the speed
  2. BSerial transmission uses no electricity, so power drops to zero
  3. CParallel transmission was banned by law
Show answer and explanation

Answer: A. In parallel transmission, arrival-time differences between wires (skew) make it hard to raise the speed

Parallel transmission requires the bits on many wires to be captured together with one clock. As speed rises, arrival-time differences caused by tiny differences in wire length (skew) become significant compared with the clock period, so the speed cannot be raised further. Serial transmission recovers the clock from the data on each pair of wires, avoiding skew, so the speed per wire can be much higher. It also reduces the number of pins and traces.

Which describes a "lane" in PCI Express?

  1. AA unit of connection with one differential pair each for transmit and receive; lanes can be bundled for more bandwidth
  2. BA single wire that alternates between sending and receiving
  3. CThe physical length of the slot into which a PCI Express device is inserted
Show answer and explanation

Answer: A. A unit of connection with one differential pair each for transmit and receive; lanes can be bundled for more bandwidth

One PCI Express lane consists of a transmit differential pair and a receive differential pair, four wires in all, and carries data in both directions simultaneously. Bundling lanes as x4 or x16 spreads data across the lanes, so bandwidth grows in proportion to the lane count. Graphics cards commonly use x16, SSDs x4 and network cards x1 to x8.

  1. What does a large CMRR (common-mode rejection ratio) of a differential amplifier mean?
  2. What is an advantage of using a differential input for an ADC?
  3. On a CAN bus with 120 Ω termination resistors at both ends, the resistance between CAN_H and CAN_L was measured with the power off. Which value is closest to that of a normal bus?

Practice I/O Interfaces & Buses questions

Key terms explained

Differential signaling
Signaling that sends opposite signals on two wires and represents 0 and 1 by their voltage difference. Noise picked up equally on both cancels, allowing fast transmission with small swings.
CMRR (common-mode rejection ratio)
The ratio of differential gain to common-mode gain in a differential amplifier. The larger it is, the better the circuit rejects noise that appears equally on both inputs.
Characteristic impedance
The ratio of voltage to current seen by a wave traveling along a transmission line, set by trace width and layer spacing. Changes along the path cause reflections.
PCI Express
A high-speed serial expansion interface connecting the CPU with GPUs, SSDs, network cards and more, also written PCIe. Lanes are bundled for bandwidth, and speed roughly doubles each generation.
Crosstalk
Noise that appears on one wire when the signal on a neighboring wire changes, mainly through capacitive coupling. Wider spacing and shield wires reduce it.

Related comparisons